A Search For Pulsations in the Optical Light Curve of the Nova ASASSN-17hx

ASASSN-17hx is a Galactic nova that was detected on UTC 2017 June 23, by the All Sky Automated Survey for SuperNovae (ASASSN), a network of robotic telescopes designed to detect transients (Stanek et al. 2017; Shappee et al. 2014). ASASSN-17hx was spectroscopically confirmed as a nova with observations from the Rozhen observatory in Bulgaria (Kurtenkov et al. 2017). In particular, it likely is a binary system with accretion from a main-sequence or giant companion onto a WD as its progenitor. While the initial spectroscopic confirmation in June identified ASASSN-17hx as a He/N nova, meaning that helium and nitrogen lines dominated the nova's optical spectrum, further spectroscopy done over the proceeding months (Guarro et al. 2017; Pavana et al. 2017; Williams & Darnley 2017), revealed increasingly strong Fe ii features. This fits the theory that most novae are hybrids of the He/N type and the Fe ii type, with the Fe ii features coming into prominence after peak brightness (Williams 2012).

According to data compiled by the American Association of Variable Star Observers (Kafka 2018), the nova first peaked on July 27th, then declined and re-brightened from August to September. It peaked again on September 16th, and experienced a third, smaller peak in early October. Multiple peaks are common in classical novae, though their physical origin is still not fully understood (Bode & Evans 2008).

The data analyzed in this paper have been acquired by the Caltech HIgh-speed Multicolor camERA (CHIMERA), a high-speed optical photometer operated in the prime focus of the Palomar 200 inch telescope (Harding et al. 2016). CHIMERA is designed to monitor objects that fluctuate on timescales ranging from milliseconds to hours. Its field of view is 5 × 5 arcmin, and when the entire CCD is read out, observations can have exposure times as low as 36 ms. Windowing to smaller fields of view, or binning pixels, allows for even lower exposure times.

CHIMERA has a dichroic that separates incoming light into a red and a blue channel, each of which is serviced by its own Electron-Multiplying (EM) CCD. This enables simultaneous observations in two distinct optical bands. Each of these channels has a separate filter mechanism, one containing SDSS u' and g', and the other containing r', i', and z'. Each channel also has a clear throughput option, with properties defined by the dichroic and CCD throughputs. See Harding et al. (2016) for throughput details.

We observed ASASSN-17hx with CHIMERA on UTC 2017 June 25, June 26, July 22, and July 23. Observation details are included in Table 1. Figure 2 shows an example image with reticles centered on ASASSN-17hx and the stars we used for differential photometry, and Figure 3 shows a light curve of the nova, including our observing dates. The June nights were mostly photometric, with some periods of light cloud cover. There were heavier periods of cloud cover during our July nights.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We performed standard debiasing and flat-fielding of the data with the PyChimera⁸ pipeline, utilizing Daophot and Pyraf (Stetson 2018). We used 0.01 s bias frames, and twilight flats, to debias and flat-field our images. Because our exposure times do not exceed 1.1 s, dark current is negligible and ignored. We performed differential photometry on the nova relative to the brightest stars in the field of view in order to normalize flux against fluctuations caused by cloud cover, atmospheric conditions, changing airmass, and other effects separate from the intrinsic nova flux. This field of view contains four stars sufficiently bright for use in differential photometry.

The baseline generated with our reference stars may be complicated by the cloud-crossing time at Palomar, combined with the 0.55–1.1 s exposure times. The cloud-crossing time is defined here as the limiting case of the average time it takes a cloud to cross CHIMERA's field of view at zenith. If the cloud-crossing time is greater than the exposure time, the fluxes of stars on opposite sides of the CHIMERA field of view are impacted at different times, therefore differential photometry may not fully cancel out the effects of cloud cover. We estimate the cloud-crossing time as the time it takes for a cirrus cloud above Palomar to cross CHIMERAs 5 × 5 arcmin field of view. Cirrus cloud speed falls within the range $\sim 45\mbox{--}71.5\,{\rm{m}}\,{{\rm{s}}}^{-1}$ (Ahrens 2006). We use the upper limit of this range in our calculation, to place a lower limit of 0.0874 s on the cloud-crossing time.

If it takes cirrus clouds roughly 0.1 s to cross CHIMERA's field of view, then the time lag of their effects on the brightness of stars on opposite sides of the field of view might be significant in 0.5 s exposures. This appears to be the case in the first night of data, where a small dip in brightness due to cloud cover is still present in the normalized nova light curve. Aside from this fluctuation, our normalized data seem largely uncorrupted by cloud cover. As shown in the next section, we can confirm this finding by comparing the time-series power spectra of our reference starlight curves to the power spectra of our nova light curves. If, on a given night, the reference stars share strong frequency peaks with that night's nova power spectrum, then these frequency peaks may be caused by cloud contamination. We plot an example of differential light curves in Figure 4.

Zoom In Zoom Out Reset image size

After differential normalization, we linearly detrended the light curves, using a least-squares linear fit and model subtraction. The light curves produced from our first night of data, both uncalibrated and normalized, are shown in Figure 4. There is still some residual variation due to changing airmass and the variation in color between the target and reference stars, on timescales longer than an hour. However, variations on these timescales do not affect our periodicity search, and the associated frequency peaks in power are excluded from our analysis.

We initially assess the quality of our differential photometry by comparing the measured noise in the resulting light curves with theoretical expectations. Both read noise and dark current are essentially negligible in each frame relative to the photon noise (of both the target and references) and scintillation noise. Whereas photon noise is simply proportional to $\sqrt{N}$, where N is the number of electron counts from the source, scintillation noise is affected by the properties of the observing telescope and observation parameters. Osborn et al. (2015) provides an estimate of a point source's fractional amplitude fluctuation σ_Y:

$$\begin{eqnarray}&&{\sigma }_{Y}^{2}={10}^{-5}{D}^{-4/3}{t}^{-1}{(\cos \gamma )}^{-3}{e}^{-2{h}_{\mathrm{obs}}/H},\end{eqnarray} \tag{ 1 }$$

where D is the telescope diameter, t is the observation exposure time, γ is the zenith distance, h_obs is the observatory altitude, and H is the atmospheric turbulence scale height, roughly 8000 m. All parameters are in SI units. Osborn et al. (2015) also note that this approximation actually underestimates the mean scintillation noise at several observatories around the world; the authors provide modified equations to more accurately estimate scintillation noise. However, these more accurate methods require measurements of the atmospheric turbulence profile as a function of altitude at the relevant observatory, preferably taken on the same night as the photometric observations. The MASS-DIMM robotic system at Palomar is capable of monitoring atmospheric turbulence, but has been inactive for several years (Thomsen et al. 2007).

For the Palomar 200 inch telescope, at an exposure time of 1.1 s and airmass 1.68 (the starting airmass of ASASSN-17hx for our 6/25 observations), σ_Y = 1.80 × 10⁻³. This value is within an order of magnitude of the true fractional noise σ/μ measured on most of our observing nights, as recorded in Table 2. Airmass of our observations on all nights ranges between 1 and 2.68, meaning that our estimate of σ_Y ranges from 8.25 × 10⁻⁴ to 3.62 × 10⁻³. This is roughly the same range spanned by the theoretical Poisson noise calculated on all nights. We therefore conclude that scintillation noise dominates for the timescales of interest, limiting our 1 s cadence data to a few mmag precision (Table 2), which is further reduced to an ∼mmag precision when binning to longer timescales (Figure 5).

Zoom In Zoom Out Reset image size

Table 2.  Nova Mean Brightness, Standard Deviation in Brightness, Mean Poisson Counting Noise, and Fractional Variations for All Observations, in Instrumental Photo-electron Units

Data Set	μ	True σ	Poisson $\sqrt{\mu }$	σ/μ	$1/\sqrt{\mu }$
Blue 6/25	3.73 × 105	2.81 × 103	6.11 × 102	7.53 × 10−3	1.64 × 10−3
Red 6/25	8.52 × 105	4.72 × 103	9.23 × 102	5.54 × 10−3	1.08 × 10−3
Blue 6/26	2.60 × 105	6.55 × 103	5.10 × 102	2.52 × 10−2	1.96 × 10−3
Red 6/26	1.12 × 106	9.25 × 103	1.06 × 103	8.26 × 10−3	9.43 × 10−4
Blue 7/22	9.52 × 105	7.91 × 103	9.76 × 102	8.31 × 10−3	1.02 × 10−3
Red 7/22	1.44 × 106	1.26 × 104	1.20 × 103	8.75 × 10−3	8.33 × 10−4
Blue 7/23	2.17 × 106	1.58 × 104	1.47 × 103	7.28 × 10−3	6.80 × 10−4
Red 7/23	1.26 × 106	1.62 × 104	1.12 × 103	1.29 × 10−2	8.93 × 10−4

Note. It should be noted that these measures are all computed for the post-differential photometry light curves. We did this to avoid including cloud cover in our noise measurements, but at the cost of adding the Poisson noise of our reference stars in quadrature to the nova Poisson noise.

Download table as:  ASCIITypeset image

We searched for periodicity in our nova light curves by calculating Lomb–Scargle periodograms (Lomb 1976; Scargle 1982; Press & Rybicki 1989; VanderPlas 2017) for each night of data and used a bootstrap method to measure the statistical significance of periodic signals. The Lomb–Scargle periodogram is an algorithm that essentially computes the Fast Fourier Transform of nonuniformly sampled time-series data. This takes into account the windowing effect caused by time delays between the FITS cubes in which CHIMERA data is saved, as well as some longer gaps in our data. Figures 6 through 9 show the computed power spectra of the nova and reference starlight curves, in both color filters and on all nights.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In order to evaluate the statistical significance of peaks in a periodogram, we used a bootstrap method to generate a roughly normal distribution of peak periodicity in our light curve data. We then used this distribution to measure the probability that a periodicity peak could arise out of random noise. To obtain this distribution, we iterated 1000 times over each normalized nova light curve. During each iteration over a time series (replaced "light curve") containing N data points, we shuffled the data and randomly sampled with replacement N times to construct a new light curve.

We then generated a Lomb–Scargle periodogram for that randomly generated light curve, recorded the periodograms peak value, and appended it to our periodicity-peak distribution. We used this distribution to compute the periodicity-peak strengths that correspond to 5% and 0.3% likelihood that the peak arose from random noise, and included these measures in plots of the true data's Lomb–Scargle periodograms, seen in Figures 6 through 9.